Abstract:

In this course an overview of different signals and systems characteristics

will be given. Mathematical methods in order of analyzing and processing

signals will be developed. Signals are analyzed in continuous and discrete linear systems.

Signal processing is performed in the time domain and in frequency domain.

Abstract:

The course will focus on several main topics: defining a basic process in machine learning; Knowing different families of machine learning paradigms, such as regression, classifier and more; Knowledge of different machine learning algorithms such as logistic regression, K-means, and DNNs.

Theme sessions:

1 Introduction: About machine learning, what types of learning exist (classification according to different types of learning), what problems can be solved.

Review: basic concepts in probability, linear algebra and optimization (finding extreme points, Lagrange multipliers, etc.).

2-4 linear regression

Logistic regression.

Regularization (1L and 2-L as an example)

Different price f?unctions (MMSE, cross-entropy)

(precision, recall) evaluation model and measures (CV, K-fold CV) methods

Practice working with the sklearn package

5 Linear SVM classifier and with kernel f?unctions

Implementation practice using sklearn

6 Non-parametric training: decision trees, kNN; Forest Random

(k-means) soft cluster + PCA, LDA, TSNE: download dimension 7

8-10 Basics of DNN

Feed-Forward network

Various activation f?unctions (linear, sigmoid, hyperbolic tangent, SoftMax, ReLu ;)

Back Propagation training

Regularization, and Out-Drop.

Model development practice using KERAS

11-12

(Optional* - may be replaced with other topics at the lecturer's discretion) Advanced architectures in machine learning

Introduction and uses of convolutional networks -CNN

Introduction to sequential models in deep learning: GRU, RNN, LSTM

13 Presentation of work 1 - review of articles

14 Presentation of work 2 - review of final project results